A fully automatic curve localization method for extracted spine

• Main Authors: Aishu Xie, Ervin Gubin Moung, Xu Zhou, Zhibang Yang
• Format: Article
• Language: English; English
• Published: Yogyakarta: Institute of Advanced Engineering and Science (IAES) 2024
• Subjects: Q1-390 Science (General); QA440-699 Geometry. Trigonometry. Topology
• Online Access: https://eprints.ums.edu.my/id/eprint/41932/1/ABSTRACT.pdf; https://eprints.ums.edu.my/id/eprint/41932/2/FULL%20TEXT.pdf; https://eprints.ums.edu.my/id/eprint/41932/; http://doi.org/10.11591/ijece.v14i4.pp4018-4033

The automation of scoliosis positioning presents a challenging and often understated task, yet it holds fundamental significance for the automated analysis of spinal morphological anomalies. This paper introduces a novel spinal curve localization model for precisely differentiating the spinal curves and identifying their concave centers. The proposed model contains three components: i) custom spine central line model, to define the spine central line as a combination of several secant line sequences with different polarities; ii) custom curve model, to classify each spinal curve into one of 11 curves types and deduce each its concave centers by several custom formulas; and iii) adapted distance transform and quadratic line fitting algorithm coupled with custom secant line segment searching strategy (DTQL-LS), to search all line segments in the spine and group consecutive line segments with identical polarity into line sequence. Experimental results show that its positioning success rate is close to 99%. Furthermore, it exhibits significant time efficiency, with the average time to process a single image being less than 30 milliseconds. Moreover, even if some image boundaries are blurred, the center of the curve can still be accurately located.